I set up Playwright tests in a TurboRepo project using Next.js, but the tests often fail — is it just me?

[chbaefront]

Summary

I configured Playwright with 4 workers to run tests that navigate through pages and verify rendered data after API responses. However, the tests sometimes pass and sometimes fail, so I’m wondering if anyone else is experiencing this issue.

My environment:

• OS: Windows
• Running tests locally
• Frontend: Next.js with Mantine UI, React Query, and Axios
• Backend: NestJS
• Database: Running in a Docker container

When I reduce the number of Playwright workers to 1 or 2, the tests usually succeed, but when I increase the number of workers, they start failing. I’m curious why this happens.

Additional information

No response

Example

No response

[IMPERIAL-HEX]

@katanazero86 Did you get the answer on this or also did you maybe tried with playwright discussions ?

[chbaefront]

I ran into a similar issue and was able to stabilize my tests with a few changes.

First, I switched to the Playwright Locator API and relied on its built-in auto-waiting instead of manually handling timing. This significantly reduced flakiness.

I also started using accessibility-based selectors (like roles and labels) when locating elements, which made the tests more resilient to UI changes.

Another important change was removing dependencies between test() cases. I made each test fully independent so they can run reliably in parallel without affecting each other.

When tests fail, I usually debug them using Playwright’s UI mode, which helps visualize what’s going on step by step.

Finally, for true E2E, validating through the API layer is ideal, but it can get quite expensive and slow. So in most cases, I mock API responses instead and focus on frontend behavior, which keeps tests faster and more stable.

Also, from my experience, Playwright E2E tests tend to be flaky by nature—especially when running in parallel—so I try to minimize external dependencies as much as possible 😅